mirror of
https://github.com/penpot/penpot.git
synced 2026-11-01 23:50:38 +00:00
Add a main RPC command, get-semantic-file, that turns a Penpot file into a small, neutral projection: pages, frames, text, links and interactions, with libraries, geometry and assets stripped. The projection is what a decision model sees, so the extractor stays model-agnostic and the deterministic checks live outside it. Add experiments/security-projection, a Node prototype that consumes the command: analyze.mjs (file -> projection -> checks -> decision model), compare.mjs (one file, several models), evaluate.mjs with a labeled set of projections, and seed.mjs, which builds real files to validate the extractor. It talks to OpenRouter decision models. Tests: backend_tests.rpc-semantic-test (3 tests, 21 assertions) and 24 Node tests across the prototype. AI-assisted-by: deepseek-v4.1-flash
15 lines
423 B
Plaintext
15 lines
423 B
Plaintext
# Copy to .env and fill in. Real environment variables win over this file.
|
|
|
|
# Penpot instance
|
|
PENPOT_BASE_URL=http://localhost:3450
|
|
|
|
# Authentication: either email + password, or an access token.
|
|
PENPOT_EMAIL=profile1@example.com
|
|
PENPOT_PASSWORD=123123
|
|
# PENPOT_ACCESS_TOKEN=
|
|
|
|
# OpenRouter
|
|
OPENROUTER_API_KEY=
|
|
OPENROUTER_MODEL=upstage/solar-decide-flash
|
|
# OPENROUTER_DECISIONS_URL=https://openrouter.ai/api/alpha/decisions
|